fs/btrfs/disk-io.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/blkdev.h>
  21 #include <linux/scatterlist.h>
  22 #include <linux/swap.h>
  23 #include <linux/radix-tree.h>
  24 #include <linux/writeback.h>
  25 #include <linux/buffer_head.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/kthread.h>
  28 #include <linux/freezer.h>
  29 #include <linux/crc32c.h>
  30 #include "compat.h"
  31 #include "ctree.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "btrfs_inode.h"
  35 #include "volumes.h"
  36 #include "print-tree.h"
  37 #include "async-thread.h"
  38 #include "locking.h"
  39 #include "tree-log.h"
  40 #include "free-space-cache.h"
  41
  42 static struct extent_io_ops btree_extent_io_ops;
  43 static void end_workqueue_fn(struct btrfs_work *work);
  44
  45 static atomic_t btrfs_bdi_num = ATOMIC_INIT(0);
  46
  47 /*
  48  * end_io_wq structs are used to do processing in task context when an IO is
  49  * complete.  This is used during reads to verify checksums, and it is used
  50  * by writes to insert metadata for new file extents after IO is complete.
  51  */
  52 struct end_io_wq {
  53         struct bio *bio;
  54         bio_end_io_t *end_io;
  55         void *private;
  56         struct btrfs_fs_info *info;
  57         int error;
  58         int metadata;
  59         struct list_head list;
  60         struct btrfs_work work;
  61 };
  62
  63 /*
  64  * async submit bios are used to offload expensive checksumming
  65  * onto the worker threads.  They checksum file and metadata bios
  66  * just before they are sent down the IO stack.
  67  */
  68 struct async_submit_bio {
  69         struct inode *inode;
  70         struct bio *bio;
  71         struct list_head list;
  72         extent_submit_bio_hook_t *submit_bio_start;
  73         extent_submit_bio_hook_t *submit_bio_done;
  74         int rw;
  75         int mirror_num;
  76         unsigned long bio_flags;
  77         struct btrfs_work work;
  78 };
  79
  80 /* These are used to set the lockdep class on the extent buffer locks.
  81  * The class is set by the readpage_end_io_hook after the buffer has
  82  * passed csum validation but before the pages are unlocked.
  83  *
  84  * The lockdep class is also set by btrfs_init_new_buffer on freshly
  85  * allocated blocks.
  86  *
  87  * The class is based on the level in the tree block, which allows lockdep
  88  * to know that lower nodes nest inside the locks of higher nodes.
  89  *
  90  * We also add a check to make sure the highest level of the tree is
  91  * the same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this
  92  * code needs update as well.
  93  */
  94 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  95 # if BTRFS_MAX_LEVEL != 8
  96 #  error
  97 # endif
  98 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
  99 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
 100         /* leaf */
 101         "btrfs-extent-00",
 102         "btrfs-extent-01",
 103         "btrfs-extent-02",
 104         "btrfs-extent-03",
 105         "btrfs-extent-04",
 106         "btrfs-extent-05",
 107         "btrfs-extent-06",
 108         "btrfs-extent-07",
 109         /* highest possible level */
 110         "btrfs-extent-08",
 111 };
 112 #endif
 113
 114 /*
 115  * extents on the btree inode are pretty simple, there's one extent
 116  * that covers the entire device
 117  */
 118 static struct extent_map *btree_get_extent(struct inode *inode,
 119                 struct page *page, size_t page_offset, u64 start, u64 len,
 120                 int create)
 121 {
 122         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 123         struct extent_map *em;
 124         int ret;
 125
 126         read_lock(&em_tree->lock);
 127         em = lookup_extent_mapping(em_tree, start, len);
 128         if (em) {
 129                 em->bdev =
 130                         BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 131                 read_unlock(&em_tree->lock);
 132                 goto out;
 133         }
 134         read_unlock(&em_tree->lock);
 135
 136         em = alloc_extent_map(GFP_NOFS);
 137         if (!em) {
 138                 em = ERR_PTR(-ENOMEM);
 139                 goto out;
 140         }
 141         em->start = 0;
 142         em->len = (u64)-1;
 143         em->block_len = (u64)-1;
 144         em->block_start = 0;
 145         em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
 146
 147         write_lock(&em_tree->lock);
 148         ret = add_extent_mapping(em_tree, em);
 149         if (ret == -EEXIST) {
 150                 u64 failed_start = em->start;
 151                 u64 failed_len = em->len;
 152
 153                 free_extent_map(em);
 154                 em = lookup_extent_mapping(em_tree, start, len);
 155                 if (em) {
 156                         ret = 0;
 157                 } else {
 158                         em = lookup_extent_mapping(em_tree, failed_start,
 159                                                    failed_len);
 160                         ret = -EIO;
 161                 }
 162         } else if (ret) {
 163                 free_extent_map(em);
 164                 em = NULL;
 165         }
 166         write_unlock(&em_tree->lock);
 167
 168         if (ret)
 169                 em = ERR_PTR(ret);
 170 out:
 171         return em;
 172 }
 173
 174 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
 175 {
 176         return crc32c(seed, data, len);
 177 }
 178
 179 void btrfs_csum_final(u32 crc, char *result)
 180 {
 181         *(__le32 *)result = ~cpu_to_le32(crc);
 182 }
 183
 184 /*
 185  * compute the csum for a btree block, and either verify it or write it
 186  * into the csum field of the block.
 187  */
 188 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
 189                            int verify)
 190 {
 191         u16 csum_size =
 192                 btrfs_super_csum_size(&root->fs_info->super_copy);
 193         char *result = NULL;
 194         unsigned long len;
 195         unsigned long cur_len;
 196         unsigned long offset = BTRFS_CSUM_SIZE;
 197         char *map_token = NULL;
 198         char *kaddr;
 199         unsigned long map_start;
 200         unsigned long map_len;
 201         int err;
 202         u32 crc = ~(u32)0;
 203         unsigned long inline_result;
 204
 205         len = buf->len - offset;
 206         while (len > 0) {
 207                 err = map_private_extent_buffer(buf, offset, 32,
 208                                         &map_token, &kaddr,
 209                                         &map_start, &map_len, KM_USER0);
 210                 if (err)
 211                         return 1;
 212                 cur_len = min(len, map_len - (offset - map_start));
 213                 crc = btrfs_csum_data(root, kaddr + offset - map_start,
 214                                       crc, cur_len);
 215                 len -= cur_len;
 216                 offset += cur_len;
 217                 unmap_extent_buffer(buf, map_token, KM_USER0);
 218         }
 219         if (csum_size > sizeof(inline_result)) {
 220                 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
 221                 if (!result)
 222                         return 1;
 223         } else {
 224                 result = (char *)&inline_result;
 225         }
 226
 227         btrfs_csum_final(crc, result);
 228
 229         if (verify) {
 230                 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
 231                         u32 val;
 232                         u32 found = 0;
 233                         memcpy(&found, result, csum_size);
 234
 235                         read_extent_buffer(buf, &val, 0, csum_size);
 236                         if (printk_ratelimit()) {
 237                                 printk(KERN_INFO "btrfs: %s checksum verify "
 238                                        "failed on %llu wanted %X found %X "
 239                                        "level %d\n",
 240                                        root->fs_info->sb->s_id,
 241                                        (unsigned long long)buf->start, val, found,
 242                                        btrfs_header_level(buf));
 243                         }
 244                         if (result != (char *)&inline_result)
 245                                 kfree(result);
 246                         return 1;
 247                 }
 248         } else {
 249                 write_extent_buffer(buf, result, 0, csum_size);
 250         }
 251         if (result != (char *)&inline_result)
 252                 kfree(result);
 253         return 0;
 254 }
 255
 256 /*
 257  * we can't consider a given block up to date unless the transid of the
 258  * block matches the transid in the parent node's pointer.  This is how we
 259  * detect blocks that either didn't get written at all or got written
 260  * in the wrong place.
 261  */
 262 static int verify_parent_transid(struct extent_io_tree *io_tree,
 263                                  struct extent_buffer *eb, u64 parent_transid)
 264 {
 265         int ret;
 266
 267         if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
 268                 return 0;
 269
 270         lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
 271         if (extent_buffer_uptodate(io_tree, eb) &&
 272             btrfs_header_generation(eb) == parent_transid) {
 273                 ret = 0;
 274                 goto out;
 275         }
 276         if (printk_ratelimit()) {
 277                 printk("parent transid verify failed on %llu wanted %llu "
 278                        "found %llu\n",
 279                        (unsigned long long)eb->start,
 280                        (unsigned long long)parent_transid,
 281                        (unsigned long long)btrfs_header_generation(eb));
 282         }
 283         ret = 1;
 284         clear_extent_buffer_uptodate(io_tree, eb);
 285 out:
 286         unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
 287                       GFP_NOFS);
 288         return ret;
 289 }
 290
 291 /*
 292  * helper to read a given tree block, doing retries as required when
 293  * the checksums don't match and we have alternate mirrors to try.
 294  */
 295 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
 296                                           struct extent_buffer *eb,
 297                                           u64 start, u64 parent_transid)
 298 {
 299         struct extent_io_tree *io_tree;
 300         int ret;
 301         int num_copies = 0;
 302         int mirror_num = 0;
 303
 304         io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
 305         while (1) {
 306                 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
 307                                                btree_get_extent, mirror_num);
 308                 if (!ret &&
 309                     !verify_parent_transid(io_tree, eb, parent_transid))
 310                         return ret;
 311
 312                 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
 313                                               eb->start, eb->len);
 314                 if (num_copies == 1)
 315                         return ret;
 316
 317                 mirror_num++;
 318                 if (mirror_num > num_copies)
 319                         return ret;
 320         }
 321         return -EIO;
 322 }
 323
 324 /*
 325  * checksum a dirty tree block before IO.  This has extra checks to make sure
 326  * we only fill in the checksum field in the first page of a multi-page block
 327  */
 328
 329 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
 330 {
 331         struct extent_io_tree *tree;
 332         u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
 333         u64 found_start;
 334         int found_level;
 335         unsigned long len;
 336         struct extent_buffer *eb;
 337         int ret;
 338
 339         tree = &BTRFS_I(page->mapping->host)->io_tree;
 340
 341         if (page->private == EXTENT_PAGE_PRIVATE)
 342                 goto out;
 343         if (!page->private)
 344                 goto out;
 345         len = page->private >> 2;
 346         WARN_ON(len == 0);
 347
 348         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 349         ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
 350                                              btrfs_header_generation(eb));
 351         BUG_ON(ret);
 352         found_start = btrfs_header_bytenr(eb);
 353         if (found_start != start) {
 354                 WARN_ON(1);
 355                 goto err;
 356         }
 357         if (eb->first_page != page) {
 358                 WARN_ON(1);
 359                 goto err;
 360         }
 361         if (!PageUptodate(page)) {
 362                 WARN_ON(1);
 363                 goto err;
 364         }
 365         found_level = btrfs_header_level(eb);
 366
 367         csum_tree_block(root, eb, 0);
 368 err:
 369         free_extent_buffer(eb);
 370 out:
 371         return 0;
 372 }
 373
 374 static int check_tree_block_fsid(struct btrfs_root *root,
 375                                  struct extent_buffer *eb)
 376 {
 377         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
 378         u8 fsid[BTRFS_UUID_SIZE];
 379         int ret = 1;
 380
 381         read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
 382                            BTRFS_FSID_SIZE);
 383         while (fs_devices) {
 384                 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
 385                         ret = 0;
 386                         break;
 387                 }
 388                 fs_devices = fs_devices->seed;
 389         }
 390         return ret;
 391 }
 392
 393 #ifdef CONFIG_DEBUG_LOCK_ALLOC
 394 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
 395 {
 396         lockdep_set_class_and_name(&eb->lock,
 397                            &btrfs_eb_class[level],
 398                            btrfs_eb_name[level]);
 399 }
 400 #endif
 401
 402 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
 403                                struct extent_state *state)
 404 {
 405         struct extent_io_tree *tree;
 406         u64 found_start;
 407         int found_level;
 408         unsigned long len;
 409         struct extent_buffer *eb;
 410         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 411         int ret = 0;
 412
 413         tree = &BTRFS_I(page->mapping->host)->io_tree;
 414         if (page->private == EXTENT_PAGE_PRIVATE)
 415                 goto out;
 416         if (!page->private)
 417                 goto out;
 418
 419         len = page->private >> 2;
 420         WARN_ON(len == 0);
 421
 422         eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
 423
 424         found_start = btrfs_header_bytenr(eb);
 425         if (found_start != start) {
 426                 if (printk_ratelimit()) {
 427                         printk(KERN_INFO "btrfs bad tree block start "
 428                                "%llu %llu\n",
 429                                (unsigned long long)found_start,
 430                                (unsigned long long)eb->start);
 431                 }
 432                 ret = -EIO;
 433                 goto err;
 434         }
 435         if (eb->first_page != page) {
 436                 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
 437                        eb->first_page->index, page->index);
 438                 WARN_ON(1);
 439                 ret = -EIO;
 440                 goto err;
 441         }
 442         if (check_tree_block_fsid(root, eb)) {
 443                 if (printk_ratelimit()) {
 444                         printk(KERN_INFO "btrfs bad fsid on block %llu\n",
 445                                (unsigned long long)eb->start);
 446                 }
 447                 ret = -EIO;
 448                 goto err;
 449         }
 450         found_level = btrfs_header_level(eb);
 451
 452         btrfs_set_buffer_lockdep_class(eb, found_level);
 453
 454         ret = csum_tree_block(root, eb, 1);
 455         if (ret)
 456                 ret = -EIO;
 457
 458         end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
 459         end = eb->start + end - 1;
 460 err:
 461         free_extent_buffer(eb);
 462 out:
 463         return ret;
 464 }
 465
 466 static void end_workqueue_bio(struct bio *bio, int err)
 467 {
 468         struct end_io_wq *end_io_wq = bio->bi_private;
 469         struct btrfs_fs_info *fs_info;
 470
 471         fs_info = end_io_wq->info;
 472         end_io_wq->error = err;
 473         end_io_wq->work.func = end_workqueue_fn;
 474         end_io_wq->work.flags = 0;
 475
 476         if (bio->bi_rw & (1 << BIO_RW)) {
 477                 if (end_io_wq->metadata)
 478                         btrfs_queue_worker(&fs_info->endio_meta_write_workers,
 479                                            &end_io_wq->work);
 480                 else
 481                         btrfs_queue_worker(&fs_info->endio_write_workers,
 482                                            &end_io_wq->work);
 483         } else {
 484                 if (end_io_wq->metadata)
 485                         btrfs_queue_worker(&fs_info->endio_meta_workers,
 486                                            &end_io_wq->work);
 487                 else
 488                         btrfs_queue_worker(&fs_info->endio_workers,
 489                                            &end_io_wq->work);
 490         }
 491 }
 492
 493 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
 494                         int metadata)
 495 {
 496         struct end_io_wq *end_io_wq;
 497         end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
 498         if (!end_io_wq)
 499                 return -ENOMEM;
 500
 501         end_io_wq->private = bio->bi_private;
 502         end_io_wq->end_io = bio->bi_end_io;
 503         end_io_wq->info = info;
 504         end_io_wq->error = 0;
 505         end_io_wq->bio = bio;
 506         end_io_wq->metadata = metadata;
 507
 508         bio->bi_private = end_io_wq;
 509         bio->bi_end_io = end_workqueue_bio;
 510         return 0;
 511 }
 512
 513 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
 514 {
 515         unsigned long limit = min_t(unsigned long,
 516                                     info->workers.max_workers,
 517                                     info->fs_devices->open_devices);
 518         return 256 * limit;
 519 }
 520
 521 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
 522 {
 523         return atomic_read(&info->nr_async_bios) >
 524                 btrfs_async_submit_limit(info);
 525 }
 526
 527 static void run_one_async_start(struct btrfs_work *work)
 528 {
 529         struct btrfs_fs_info *fs_info;
 530         struct async_submit_bio *async;
 531
 532         async = container_of(work, struct  async_submit_bio, work);
 533         fs_info = BTRFS_I(async->inode)->root->fs_info;
 534         async->submit_bio_start(async->inode, async->rw, async->bio,
 535                                async->mirror_num, async->bio_flags);
 536 }
 537
 538 static void run_one_async_done(struct btrfs_work *work)
 539 {
 540         struct btrfs_fs_info *fs_info;
 541         struct async_submit_bio *async;
 542         int limit;
 543
 544         async = container_of(work, struct  async_submit_bio, work);
 545         fs_info = BTRFS_I(async->inode)->root->fs_info;
 546
 547         limit = btrfs_async_submit_limit(fs_info);
 548         limit = limit * 2 / 3;
 549
 550         atomic_dec(&fs_info->nr_async_submits);
 551
 552         if (atomic_read(&fs_info->nr_async_submits) < limit &&
 553             waitqueue_active(&fs_info->async_submit_wait))
 554                 wake_up(&fs_info->async_submit_wait);
 555
 556         async->submit_bio_done(async->inode, async->rw, async->bio,
 557                                async->mirror_num, async->bio_flags);
 558 }
 559
 560 static void run_one_async_free(struct btrfs_work *work)
 561 {
 562         struct async_submit_bio *async;
 563
 564         async = container_of(work, struct  async_submit_bio, work);
 565         kfree(async);
 566 }
 567
 568 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
 569                         int rw, struct bio *bio, int mirror_num,
 570                         unsigned long bio_flags,
 571                         extent_submit_bio_hook_t *submit_bio_start,
 572                         extent_submit_bio_hook_t *submit_bio_done)
 573 {
 574         struct async_submit_bio *async;
 575
 576         async = kmalloc(sizeof(*async), GFP_NOFS);
 577         if (!async)
 578                 return -ENOMEM;
 579
 580         async->inode = inode;
 581         async->rw = rw;
 582         async->bio = bio;
 583         async->mirror_num = mirror_num;
 584         async->submit_bio_start = submit_bio_start;
 585         async->submit_bio_done = submit_bio_done;
 586
 587         async->work.func = run_one_async_start;
 588         async->work.ordered_func = run_one_async_done;
 589         async->work.ordered_free = run_one_async_free;
 590
 591         async->work.flags = 0;
 592         async->bio_flags = bio_flags;
 593
 594         atomic_inc(&fs_info->nr_async_submits);
 595
 596         if (rw & (1 << BIO_RW_SYNCIO))
 597                 btrfs_set_work_high_prio(&async->work);
 598
 599         btrfs_queue_worker(&fs_info->workers, &async->work);
 600
 601         while (atomic_read(&fs_info->async_submit_draining) &&
 602               atomic_read(&fs_info->nr_async_submits)) {
 603                 wait_event(fs_info->async_submit_wait,
 604                            (atomic_read(&fs_info->nr_async_submits) == 0));
 605         }
 606
 607         return 0;
 608 }
 609
 610 static int btree_csum_one_bio(struct bio *bio)
 611 {
 612         struct bio_vec *bvec = bio->bi_io_vec;
 613         int bio_index = 0;
 614         struct btrfs_root *root;
 615
 616         WARN_ON(bio->bi_vcnt <= 0);
 617         while (bio_index < bio->bi_vcnt) {
 618                 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
 619                 csum_dirty_buffer(root, bvec->bv_page);
 620                 bio_index++;
 621                 bvec++;
 622         }
 623         return 0;
 624 }
 625
 626 static int __btree_submit_bio_start(struct inode *inode, int rw,
 627                                     struct bio *bio, int mirror_num,
 628                                     unsigned long bio_flags)
 629 {
 630         /*
 631          * when we're called for a write, we're already in the async
 632          * submission context.  Just jump into btrfs_map_bio
 633          */
 634         btree_csum_one_bio(bio);
 635         return 0;
 636 }
 637
 638 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
 639                                  int mirror_num, unsigned long bio_flags)
 640 {
 641         /*
 642          * when we're called for a write, we're already in the async
 643          * submission context.  Just jump into btrfs_map_bio
 644          */
 645         return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
 646 }
 647
 648 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
 649                                  int mirror_num, unsigned long bio_flags)
 650 {
 651         int ret;
 652
 653         ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
 654                                           bio, 1);
 655         BUG_ON(ret);
 656
 657         if (!(rw & (1 << BIO_RW))) {
 658                 /*
 659                  * called for a read, do the setup so that checksum validation
 660                  * can happen in the async kernel threads
 661                  */
 662                 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
 663                                      mirror_num, 0);
 664         }
 665
 666         /*
 667          * kthread helpers are used to submit writes so that checksumming
 668          * can happen in parallel across all CPUs
 669          */
 670         return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
 671                                    inode, rw, bio, mirror_num, 0,
 672                                    __btree_submit_bio_start,
 673                                    __btree_submit_bio_done);
 674 }
 675
 676 static int btree_writepage(struct page *page, struct writeback_control *wbc)
 677 {
 678         struct extent_io_tree *tree;
 679         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
 680         struct extent_buffer *eb;
 681         int was_dirty;
 682
 683         tree = &BTRFS_I(page->mapping->host)->io_tree;
 684         if (!(current->flags & PF_MEMALLOC)) {
 685                 return extent_write_full_page(tree, page,
 686                                               btree_get_extent, wbc);
 687         }
 688
 689         redirty_page_for_writepage(wbc, page);
 690         eb = btrfs_find_tree_block(root, page_offset(page),
 691                                       PAGE_CACHE_SIZE);
 692         WARN_ON(!eb);
 693
 694         was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
 695         if (!was_dirty) {
 696                 spin_lock(&root->fs_info->delalloc_lock);
 697                 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
 698                 spin_unlock(&root->fs_info->delalloc_lock);
 699         }
 700         free_extent_buffer(eb);
 701
 702         unlock_page(page);
 703         return 0;
 704 }
 705
 706 static int btree_writepages(struct address_space *mapping,
 707                             struct writeback_control *wbc)
 708 {
 709         struct extent_io_tree *tree;
 710         tree = &BTRFS_I(mapping->host)->io_tree;
 711         if (wbc->sync_mode == WB_SYNC_NONE) {
 712                 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
 713                 u64 num_dirty;
 714                 unsigned long thresh = 32 * 1024 * 1024;
 715
 716                 if (wbc->for_kupdate)
 717                         return 0;
 718
 719                 /* this is a bit racy, but that's ok */
 720                 num_dirty = root->fs_info->dirty_metadata_bytes;
 721                 if (num_dirty < thresh)
 722                         return 0;
 723         }
 724         return extent_writepages(tree, mapping, btree_get_extent, wbc);
 725 }
 726
 727 static int btree_readpage(struct file *file, struct page *page)
 728 {
 729         struct extent_io_tree *tree;
 730         tree = &BTRFS_I(page->mapping->host)->io_tree;
 731         return extent_read_full_page(tree, page, btree_get_extent);
 732 }
 733
 734 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
 735 {
 736         struct extent_io_tree *tree;
 737         struct extent_map_tree *map;
 738         int ret;
 739
 740         if (PageWriteback(page) || PageDirty(page))
 741                 return 0;
 742
 743         tree = &BTRFS_I(page->mapping->host)->io_tree;
 744         map = &BTRFS_I(page->mapping->host)->extent_tree;
 745
 746         ret = try_release_extent_state(map, tree, page, gfp_flags);
 747         if (!ret)
 748                 return 0;
 749
 750         ret = try_release_extent_buffer(tree, page);
 751         if (ret == 1) {
 752                 ClearPagePrivate(page);
 753                 set_page_private(page, 0);
 754                 page_cache_release(page);
 755         }
 756
 757         return ret;
 758 }
 759
 760 static void btree_invalidatepage(struct page *page, unsigned long offset)
 761 {
 762         struct extent_io_tree *tree;
 763         tree = &BTRFS_I(page->mapping->host)->io_tree;
 764         extent_invalidatepage(tree, page, offset);
 765         btree_releasepage(page, GFP_NOFS);
 766         if (PagePrivate(page)) {
 767                 printk(KERN_WARNING "btrfs warning page private not zero "
 768                        "on page %llu\n", (unsigned long long)page_offset(page));
 769                 ClearPagePrivate(page);
 770                 set_page_private(page, 0);
 771                 page_cache_release(page);
 772         }
 773 }
 774
 775 static struct address_space_operations btree_aops = {
 776         .readpage       = btree_readpage,
 777         .writepage      = btree_writepage,
 778         .writepages     = btree_writepages,
 779         .releasepage    = btree_releasepage,
 780         .invalidatepage = btree_invalidatepage,
 781         .sync_page      = block_sync_page,
 782 };
 783
 784 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
 785                          u64 parent_transid)
 786 {
 787         struct extent_buffer *buf = NULL;
 788         struct inode *btree_inode = root->fs_info->btree_inode;
 789         int ret = 0;
 790
 791         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 792         if (!buf)
 793                 return 0;
 794         read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
 795                                  buf, 0, 0, btree_get_extent, 0);
 796         free_extent_buffer(buf);
 797         return ret;
 798 }
 799
 800 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
 801                                             u64 bytenr, u32 blocksize)
 802 {
 803         struct inode *btree_inode = root->fs_info->btree_inode;
 804         struct extent_buffer *eb;
 805         eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 806                                 bytenr, blocksize, GFP_NOFS);
 807         return eb;
 808 }
 809
 810 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
 811                                                  u64 bytenr, u32 blocksize)
 812 {
 813         struct inode *btree_inode = root->fs_info->btree_inode;
 814         struct extent_buffer *eb;
 815
 816         eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
 817                                  bytenr, blocksize, NULL, GFP_NOFS);
 818         return eb;
 819 }
 820
 821
 822 int btrfs_write_tree_block(struct extent_buffer *buf)
 823 {
 824         return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
 825                                       buf->start + buf->len - 1, WB_SYNC_ALL);
 826 }
 827
 828 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
 829 {
 830         return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
 831                                   buf->start, buf->start + buf->len - 1);
 832 }
 833
 834 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
 835                                       u32 blocksize, u64 parent_transid)
 836 {
 837         struct extent_buffer *buf = NULL;
 838         struct inode *btree_inode = root->fs_info->btree_inode;
 839         struct extent_io_tree *io_tree;
 840         int ret;
 841
 842         io_tree = &BTRFS_I(btree_inode)->io_tree;
 843
 844         buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
 845         if (!buf)
 846                 return NULL;
 847
 848         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
 849
 850         if (ret == 0)
 851                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
 852         return buf;
 853
 854 }
 855
 856 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
 857                      struct extent_buffer *buf)
 858 {
 859         struct inode *btree_inode = root->fs_info->btree_inode;
 860         if (btrfs_header_generation(buf) ==
 861             root->fs_info->running_transaction->transid) {
 862                 btrfs_assert_tree_locked(buf);
 863
 864                 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
 865                         spin_lock(&root->fs_info->delalloc_lock);
 866                         if (root->fs_info->dirty_metadata_bytes >= buf->len)
 867                                 root->fs_info->dirty_metadata_bytes -= buf->len;
 868                         else
 869                                 WARN_ON(1);
 870                         spin_unlock(&root->fs_info->delalloc_lock);
 871                 }
 872
 873                 /* ugh, clear_extent_buffer_dirty needs to lock the page */
 874                 btrfs_set_lock_blocking(buf);
 875                 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
 876                                           buf);
 877         }
 878         return 0;
 879 }
 880
 881 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
 882                         u32 stripesize, struct btrfs_root *root,
 883                         struct btrfs_fs_info *fs_info,
 884                         u64 objectid)
 885 {
 886         root->node = NULL;
 887         root->commit_root = NULL;
 888         root->sectorsize = sectorsize;
 889         root->nodesize = nodesize;
 890         root->leafsize = leafsize;
 891         root->stripesize = stripesize;
 892         root->ref_cows = 0;
 893         root->track_dirty = 0;
 894
 895         root->fs_info = fs_info;
 896         root->objectid = objectid;
 897         root->last_trans = 0;
 898         root->highest_objectid = 0;
 899         root->name = NULL;
 900         root->in_sysfs = 0;
 901         root->inode_tree.rb_node = NULL;
 902
 903         INIT_LIST_HEAD(&root->dirty_list);
 904         INIT_LIST_HEAD(&root->orphan_list);
 905         INIT_LIST_HEAD(&root->root_list);
 906         spin_lock_init(&root->node_lock);
 907         spin_lock_init(&root->list_lock);
 908         spin_lock_init(&root->inode_lock);
 909         mutex_init(&root->objectid_mutex);
 910         mutex_init(&root->log_mutex);
 911         init_waitqueue_head(&root->log_writer_wait);
 912         init_waitqueue_head(&root->log_commit_wait[0]);
 913         init_waitqueue_head(&root->log_commit_wait[1]);
 914         atomic_set(&root->log_commit[0], 0);
 915         atomic_set(&root->log_commit[1], 0);
 916         atomic_set(&root->log_writers, 0);
 917         root->log_batch = 0;
 918         root->log_transid = 0;
 919         extent_io_tree_init(&root->dirty_log_pages,
 920                              fs_info->btree_inode->i_mapping, GFP_NOFS);
 921
 922         memset(&root->root_key, 0, sizeof(root->root_key));
 923         memset(&root->root_item, 0, sizeof(root->root_item));
 924         memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
 925         memset(&root->root_kobj, 0, sizeof(root->root_kobj));
 926         root->defrag_trans_start = fs_info->generation;
 927         init_completion(&root->kobj_unregister);
 928         root->defrag_running = 0;
 929         root->defrag_level = 0;
 930         root->root_key.objectid = objectid;
 931         root->anon_super.s_root = NULL;
 932         root->anon_super.s_dev = 0;
 933         INIT_LIST_HEAD(&root->anon_super.s_list);
 934         INIT_LIST_HEAD(&root->anon_super.s_instances);
 935         init_rwsem(&root->anon_super.s_umount);
 936
 937         return 0;
 938 }
 939
 940 static int find_and_setup_root(struct btrfs_root *tree_root,
 941                                struct btrfs_fs_info *fs_info,
 942                                u64 objectid,
 943                                struct btrfs_root *root)
 944 {
 945         int ret;
 946         u32 blocksize;
 947         u64 generation;
 948
 949         __setup_root(tree_root->nodesize, tree_root->leafsize,
 950                      tree_root->sectorsize, tree_root->stripesize,
 951                      root, fs_info, objectid);
 952         ret = btrfs_find_last_root(tree_root, objectid,
 953                                    &root->root_item, &root->root_key);
 954         BUG_ON(ret);
 955
 956         generation = btrfs_root_generation(&root->root_item);
 957         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
 958         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
 959                                      blocksize, generation);
 960         root->commit_root = btrfs_root_node(root);
 961         BUG_ON(!root->node);
 962         return 0;
 963 }
 964
 965 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
 966                              struct btrfs_fs_info *fs_info)
 967 {
 968         struct extent_buffer *eb;
 969         struct btrfs_root *log_root_tree = fs_info->log_root_tree;
 970         u64 start = 0;
 971         u64 end = 0;
 972         int ret;
 973
 974         if (!log_root_tree)
 975                 return 0;
 976
 977         while (1) {
 978                 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
 979                                     0, &start, &end, EXTENT_DIRTY);
 980                 if (ret)
 981                         break;
 982
 983                 clear_extent_dirty(&log_root_tree->dirty_log_pages,
 984                                    start, end, GFP_NOFS);
 985         }
 986         eb = fs_info->log_root_tree->node;
 987
 988         WARN_ON(btrfs_header_level(eb) != 0);
 989         WARN_ON(btrfs_header_nritems(eb) != 0);
 990
 991         ret = btrfs_free_reserved_extent(fs_info->tree_root,
 992                                 eb->start, eb->len);
 993         BUG_ON(ret);
 994
 995         free_extent_buffer(eb);
 996         kfree(fs_info->log_root_tree);
 997         fs_info->log_root_tree = NULL;
 998         return 0;
 999 }
1000
1001 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1002                                          struct btrfs_fs_info *fs_info)
1003 {
1004         struct btrfs_root *root;
1005         struct btrfs_root *tree_root = fs_info->tree_root;
1006         struct extent_buffer *leaf;
1007
1008         root = kzalloc(sizeof(*root), GFP_NOFS);
1009         if (!root)
1010                 return ERR_PTR(-ENOMEM);
1011
1012         __setup_root(tree_root->nodesize, tree_root->leafsize,
1013                      tree_root->sectorsize, tree_root->stripesize,
1014                      root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1015
1016         root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1017         root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1018         root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1019         /*
1020          * log trees do not get reference counted because they go away
1021          * before a real commit is actually done.  They do store pointers
1022          * to file data extents, and those reference counts still get
1023          * updated (along with back refs to the log tree).
1024          */
1025         root->ref_cows = 0;
1026
1027         leaf = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
1028                                       BTRFS_TREE_LOG_OBJECTID, NULL, 0, 0, 0);
1029         if (IS_ERR(leaf)) {
1030                 kfree(root);
1031                 return ERR_CAST(leaf);
1032         }
1033
1034         memset_extent_buffer(leaf, 0, 0, sizeof(struct btrfs_header));
1035         btrfs_set_header_bytenr(leaf, leaf->start);
1036         btrfs_set_header_generation(leaf, trans->transid);
1037         btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
1038         btrfs_set_header_owner(leaf, BTRFS_TREE_LOG_OBJECTID);
1039         root->node = leaf;
1040
1041         write_extent_buffer(root->node, root->fs_info->fsid,
1042                             (unsigned long)btrfs_header_fsid(root->node),
1043                             BTRFS_FSID_SIZE);
1044         btrfs_mark_buffer_dirty(root->node);
1045         btrfs_tree_unlock(root->node);
1046         return root;
1047 }
1048
1049 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1050                              struct btrfs_fs_info *fs_info)
1051 {
1052         struct btrfs_root *log_root;
1053
1054         log_root = alloc_log_tree(trans, fs_info);
1055         if (IS_ERR(log_root))
1056                 return PTR_ERR(log_root);
1057         WARN_ON(fs_info->log_root_tree);
1058         fs_info->log_root_tree = log_root;
1059         return 0;
1060 }
1061
1062 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1063                        struct btrfs_root *root)
1064 {
1065         struct btrfs_root *log_root;
1066         struct btrfs_inode_item *inode_item;
1067
1068         log_root = alloc_log_tree(trans, root->fs_info);
1069         if (IS_ERR(log_root))
1070                 return PTR_ERR(log_root);
1071
1072         log_root->last_trans = trans->transid;
1073         log_root->root_key.offset = root->root_key.objectid;
1074
1075         inode_item = &log_root->root_item.inode;
1076         inode_item->generation = cpu_to_le64(1);
1077         inode_item->size = cpu_to_le64(3);
1078         inode_item->nlink = cpu_to_le32(1);
1079         inode_item->nbytes = cpu_to_le64(root->leafsize);
1080         inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1081
1082         btrfs_set_root_node(&log_root->root_item, log_root->node);
1083
1084         WARN_ON(root->log_root);
1085         root->log_root = log_root;
1086         root->log_transid = 0;
1087         return 0;
1088 }
1089
1090 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1091                                                struct btrfs_key *location)
1092 {
1093         struct btrfs_root *root;
1094         struct btrfs_fs_info *fs_info = tree_root->fs_info;
1095         struct btrfs_path *path;
1096         struct extent_buffer *l;
1097         u64 generation;
1098         u32 blocksize;
1099         int ret = 0;
1100
1101         root = kzalloc(sizeof(*root), GFP_NOFS);
1102         if (!root)
1103                 return ERR_PTR(-ENOMEM);
1104         if (location->offset == (u64)-1) {
1105                 ret = find_and_setup_root(tree_root, fs_info,
1106                                           location->objectid, root);
1107                 if (ret) {
1108                         kfree(root);
1109                         return ERR_PTR(ret);
1110                 }
1111                 goto out;
1112         }
1113
1114         __setup_root(tree_root->nodesize, tree_root->leafsize,
1115                      tree_root->sectorsize, tree_root->stripesize,
1116                      root, fs_info, location->objectid);
1117
1118         path = btrfs_alloc_path();
1119         BUG_ON(!path);
1120         ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1121         if (ret == 0) {
1122                 l = path->nodes[0];
1123                 read_extent_buffer(l, &root->root_item,
1124                                 btrfs_item_ptr_offset(l, path->slots[0]),
1125                                 sizeof(root->root_item));
1126                 memcpy(&root->root_key, location, sizeof(*location));
1127         }
1128         btrfs_free_path(path);
1129         if (ret) {
1130                 if (ret > 0)
1131                         ret = -ENOENT;
1132                 return ERR_PTR(ret);
1133         }
1134
1135         generation = btrfs_root_generation(&root->root_item);
1136         blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1137         root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1138                                      blocksize, generation);
1139         root->commit_root = btrfs_root_node(root);
1140         BUG_ON(!root->node);
1141 out:
1142         if (location->objectid != BTRFS_TREE_LOG_OBJECTID)
1143                 root->ref_cows = 1;
1144
1145         return root;
1146 }
1147
1148 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1149                                         u64 root_objectid)
1150 {
1151         struct btrfs_root *root;
1152
1153         if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1154                 return fs_info->tree_root;
1155         if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1156                 return fs_info->extent_root;
1157
1158         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1159                                  (unsigned long)root_objectid);
1160         return root;
1161 }
1162
1163 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1164                                               struct btrfs_key *location)
1165 {
1166         struct btrfs_root *root;
1167         int ret;
1168
1169         if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1170                 return fs_info->tree_root;
1171         if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1172                 return fs_info->extent_root;
1173         if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1174                 return fs_info->chunk_root;
1175         if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1176                 return fs_info->dev_root;
1177         if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1178                 return fs_info->csum_root;
1179
1180         root = radix_tree_lookup(&fs_info->fs_roots_radix,
1181                                  (unsigned long)location->objectid);
1182         if (root)
1183                 return root;
1184
1185         root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1186         if (IS_ERR(root))
1187                 return root;
1188
1189         set_anon_super(&root->anon_super, NULL);
1190
1191         ret = radix_tree_insert(&fs_info->fs_roots_radix,
1192                                 (unsigned long)root->root_key.objectid,
1193                                 root);
1194         if (ret) {
1195                 free_extent_buffer(root->node);
1196                 kfree(root);
1197                 return ERR_PTR(ret);
1198         }
1199         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1200                 ret = btrfs_find_dead_roots(fs_info->tree_root,
1201                                             root->root_key.objectid);
1202                 BUG_ON(ret);
1203                 btrfs_orphan_cleanup(root);
1204         }
1205         return root;
1206 }
1207
1208 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1209                                       struct btrfs_key *location,
1210                                       const char *name, int namelen)
1211 {
1212         struct btrfs_root *root;
1213         int ret;
1214
1215         root = btrfs_read_fs_root_no_name(fs_info, location);
1216         if (!root)
1217                 return NULL;
1218
1219         if (root->in_sysfs)
1220                 return root;
1221
1222         ret = btrfs_set_root_name(root, name, namelen);
1223         if (ret) {
1224                 free_extent_buffer(root->node);
1225                 kfree(root);
1226                 return ERR_PTR(ret);
1227         }
1228 #if 0
1229         ret = btrfs_sysfs_add_root(root);
1230         if (ret) {
1231                 free_extent_buffer(root->node);
1232                 kfree(root->name);
1233                 kfree(root);
1234                 return ERR_PTR(ret);
1235         }
1236 #endif
1237         root->in_sysfs = 1;
1238         return root;
1239 }
1240
1241 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1242 {
1243         struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1244         int ret = 0;
1245         struct btrfs_device *device;
1246         struct backing_dev_info *bdi;
1247
1248         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1249                 if (!device->bdev)
1250                         continue;
1251                 bdi = blk_get_backing_dev_info(device->bdev);
1252                 if (bdi && bdi_congested(bdi, bdi_bits)) {
1253                         ret = 1;
1254                         break;
1255                 }
1256         }
1257         return ret;
1258 }
1259
1260 /*
1261  * this unplugs every device on the box, and it is only used when page
1262  * is null
1263  */
1264 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1265 {
1266         struct btrfs_device *device;
1267         struct btrfs_fs_info *info;
1268
1269         info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1270         list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1271                 if (!device->bdev)
1272                         continue;
1273
1274                 bdi = blk_get_backing_dev_info(device->bdev);
1275                 if (bdi->unplug_io_fn)
1276                         bdi->unplug_io_fn(bdi, page);
1277         }
1278 }
1279
1280 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1281 {
1282         struct inode *inode;
1283         struct extent_map_tree *em_tree;
1284         struct extent_map *em;
1285         struct address_space *mapping;
1286         u64 offset;
1287
1288         /* the generic O_DIRECT read code does this */
1289         if (1 || !page) {
1290                 __unplug_io_fn(bdi, page);
1291                 return;
1292         }
1293
1294         /*
1295          * page->mapping may change at any time.  Get a consistent copy
1296          * and use that for everything below
1297          */
1298         smp_mb();
1299         mapping = page->mapping;
1300         if (!mapping)
1301                 return;
1302
1303         inode = mapping->host;
1304
1305         /*
1306          * don't do the expensive searching for a small number of
1307          * devices
1308          */
1309         if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1310                 __unplug_io_fn(bdi, page);
1311                 return;
1312         }
1313
1314         offset = page_offset(page);
1315
1316         em_tree = &BTRFS_I(inode)->extent_tree;
1317         read_lock(&em_tree->lock);
1318         em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1319         read_unlock(&em_tree->lock);
1320         if (!em) {
1321                 __unplug_io_fn(bdi, page);
1322                 return;
1323         }
1324
1325         if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1326                 free_extent_map(em);
1327                 __unplug_io_fn(bdi, page);
1328                 return;
1329         }
1330         offset = offset - em->start;
1331         btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1332                           em->block_start + offset, page);
1333         free_extent_map(em);
1334 }
1335
1336 /*
1337  * If this fails, caller must call bdi_destroy() to get rid of the
1338  * bdi again.
1339  */
1340 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1341 {
1342         int err;
1343
1344         bdi->capabilities = BDI_CAP_MAP_COPY;
1345         err = bdi_init(bdi);
1346         if (err)
1347                 return err;
1348
1349         err = bdi_register(bdi, NULL, "btrfs-%d",
1350                                 atomic_inc_return(&btrfs_bdi_num));
1351         if (err)
1352                 return err;
1353
1354         bdi->ra_pages   = default_backing_dev_info.ra_pages;
1355         bdi->unplug_io_fn       = btrfs_unplug_io_fn;
1356         bdi->unplug_io_data     = info;
1357         bdi->congested_fn       = btrfs_congested_fn;
1358         bdi->congested_data     = info;
1359         return 0;
1360 }
1361
1362 static int bio_ready_for_csum(struct bio *bio)
1363 {
1364         u64 length = 0;
1365         u64 buf_len = 0;
1366         u64 start = 0;
1367         struct page *page;
1368         struct extent_io_tree *io_tree = NULL;
1369         struct btrfs_fs_info *info = NULL;
1370         struct bio_vec *bvec;
1371         int i;
1372         int ret;
1373
1374         bio_for_each_segment(bvec, bio, i) {
1375                 page = bvec->bv_page;
1376                 if (page->private == EXTENT_PAGE_PRIVATE) {
1377                         length += bvec->bv_len;
1378                         continue;
1379                 }
1380                 if (!page->private) {
1381                         length += bvec->bv_len;
1382                         continue;
1383                 }
1384                 length = bvec->bv_len;
1385                 buf_len = page->private >> 2;
1386                 start = page_offset(page) + bvec->bv_offset;
1387                 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1388                 info = BTRFS_I(page->mapping->host)->root->fs_info;
1389         }
1390         /* are we fully contained in this bio? */
1391         if (buf_len <= length)
1392                 return 1;
1393
1394         ret = extent_range_uptodate(io_tree, start + length,
1395                                     start + buf_len - 1);
1396         return ret;
1397 }
1398
1399 /*
1400  * called by the kthread helper functions to finally call the bio end_io
1401  * functions.  This is where read checksum verification actually happens
1402  */
1403 static void end_workqueue_fn(struct btrfs_work *work)
1404 {
1405         struct bio *bio;
1406         struct end_io_wq *end_io_wq;
1407         struct btrfs_fs_info *fs_info;
1408         int error;
1409
1410         end_io_wq = container_of(work, struct end_io_wq, work);
1411         bio = end_io_wq->bio;
1412         fs_info = end_io_wq->info;
1413
1414         /* metadata bio reads are special because the whole tree block must
1415          * be checksummed at once.  This makes sure the entire block is in
1416          * ram and up to date before trying to verify things.  For
1417          * blocksize <= pagesize, it is basically a noop
1418          */
1419         if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1420             !bio_ready_for_csum(bio)) {
1421                 btrfs_queue_worker(&fs_info->endio_meta_workers,
1422                                    &end_io_wq->work);
1423                 return;
1424         }
1425         error = end_io_wq->error;
1426         bio->bi_private = end_io_wq->private;
1427         bio->bi_end_io = end_io_wq->end_io;
1428         kfree(end_io_wq);
1429         bio_endio(bio, error);
1430 }
1431
1432 static int cleaner_kthread(void *arg)
1433 {
1434         struct btrfs_root *root = arg;
1435
1436         do {
1437                 smp_mb();
1438                 if (root->fs_info->closing)
1439                         break;
1440
1441                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1442                 mutex_lock(&root->fs_info->cleaner_mutex);
1443                 btrfs_clean_old_snapshots(root);
1444                 mutex_unlock(&root->fs_info->cleaner_mutex);
1445
1446                 if (freezing(current)) {
1447                         refrigerator();
1448                 } else {
1449                         smp_mb();
1450                         if (root->fs_info->closing)
1451                                 break;
1452                         set_current_state(TASK_INTERRUPTIBLE);
1453                         schedule();
1454                         __set_current_state(TASK_RUNNING);
1455                 }
1456         } while (!kthread_should_stop());
1457         return 0;
1458 }
1459
1460 static int transaction_kthread(void *arg)
1461 {
1462         struct btrfs_root *root = arg;
1463         struct btrfs_trans_handle *trans;
1464         struct btrfs_transaction *cur;
1465         unsigned long now;
1466         unsigned long delay;
1467         int ret;
1468
1469         do {
1470                 smp_mb();
1471                 if (root->fs_info->closing)
1472                         break;
1473
1474                 delay = HZ * 30;
1475                 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1476                 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1477
1478                 mutex_lock(&root->fs_info->trans_mutex);
1479                 cur = root->fs_info->running_transaction;
1480                 if (!cur) {
1481                         mutex_unlock(&root->fs_info->trans_mutex);
1482                         goto sleep;
1483                 }
1484
1485                 now = get_seconds();
1486                 if (now < cur->start_time || now - cur->start_time < 30) {
1487                         mutex_unlock(&root->fs_info->trans_mutex);
1488                         delay = HZ * 5;
1489                         goto sleep;
1490                 }
1491                 mutex_unlock(&root->fs_info->trans_mutex);
1492                 trans = btrfs_start_transaction(root, 1);
1493                 ret = btrfs_commit_transaction(trans, root);
1494
1495 sleep:
1496                 wake_up_process(root->fs_info->cleaner_kthread);
1497                 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1498
1499                 if (freezing(current)) {
1500                         refrigerator();
1501                 } else {
1502                         if (root->fs_info->closing)
1503                                 break;
1504                         set_current_state(TASK_INTERRUPTIBLE);
1505                         schedule_timeout(delay);
1506                         __set_current_state(TASK_RUNNING);
1507                 }
1508         } while (!kthread_should_stop());
1509         return 0;
1510 }
1511
1512 struct btrfs_root *open_ctree(struct super_block *sb,
1513                               struct btrfs_fs_devices *fs_devices,
1514                               char *options)
1515 {
1516         u32 sectorsize;
1517         u32 nodesize;
1518         u32 leafsize;
1519         u32 blocksize;
1520         u32 stripesize;
1521         u64 generation;
1522         u64 features;
1523         struct btrfs_key location;
1524         struct buffer_head *bh;
1525         struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1526                                                  GFP_NOFS);
1527         struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1528                                                  GFP_NOFS);
1529         struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1530                                                GFP_NOFS);
1531         struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1532                                                 GFP_NOFS);
1533         struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1534                                                 GFP_NOFS);
1535         struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1536                                               GFP_NOFS);
1537         struct btrfs_root *log_tree_root;
1538
1539         int ret;
1540         int err = -EINVAL;
1541
1542         struct btrfs_super_block *disk_super;
1543
1544         if (!extent_root || !tree_root || !fs_info ||
1545             !chunk_root || !dev_root || !csum_root) {
1546                 err = -ENOMEM;
1547                 goto fail;
1548         }
1549         INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1550         INIT_LIST_HEAD(&fs_info->trans_list);
1551         INIT_LIST_HEAD(&fs_info->dead_roots);
1552         INIT_LIST_HEAD(&fs_info->hashers);
1553         INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1554         INIT_LIST_HEAD(&fs_info->ordered_operations);
1555         INIT_LIST_HEAD(&fs_info->caching_block_groups);
1556         spin_lock_init(&fs_info->delalloc_lock);
1557         spin_lock_init(&fs_info->new_trans_lock);
1558         spin_lock_init(&fs_info->ref_cache_lock);
1559
1560         init_completion(&fs_info->kobj_unregister);
1561         fs_info->tree_root = tree_root;
1562         fs_info->extent_root = extent_root;
1563         fs_info->csum_root = csum_root;
1564         fs_info->chunk_root = chunk_root;
1565         fs_info->dev_root = dev_root;
1566         fs_info->fs_devices = fs_devices;
1567         INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1568         INIT_LIST_HEAD(&fs_info->space_info);
1569         btrfs_mapping_init(&fs_info->mapping_tree);
1570         atomic_set(&fs_info->nr_async_submits, 0);
1571         atomic_set(&fs_info->async_delalloc_pages, 0);
1572         atomic_set(&fs_info->async_submit_draining, 0);
1573         atomic_set(&fs_info->nr_async_bios, 0);
1574         fs_info->sb = sb;
1575         fs_info->max_extent = (u64)-1;
1576         fs_info->max_inline = 8192 * 1024;
1577         if (setup_bdi(fs_info, &fs_info->bdi))
1578                 goto fail_bdi;
1579         fs_info->btree_inode = new_inode(sb);
1580         fs_info->btree_inode->i_ino = 1;
1581         fs_info->btree_inode->i_nlink = 1;
1582         fs_info->metadata_ratio = 8;
1583
1584         fs_info->thread_pool_size = min_t(unsigned long,
1585                                           num_online_cpus() + 2, 8);
1586
1587         INIT_LIST_HEAD(&fs_info->ordered_extents);
1588         spin_lock_init(&fs_info->ordered_extent_lock);
1589
1590         sb->s_blocksize = 4096;
1591         sb->s_blocksize_bits = blksize_bits(4096);
1592
1593         /*
1594          * we set the i_size on the btree inode to the max possible int.
1595          * the real end of the address space is determined by all of
1596          * the devices in the system
1597          */
1598         fs_info->btree_inode->i_size = OFFSET_MAX;
1599         fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1600         fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1601
1602         RB_CLEAR_NODE(&BTRFS_I(fs_info->btree_inode)->rb_node);
1603         extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1604                              fs_info->btree_inode->i_mapping,
1605                              GFP_NOFS);
1606         extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1607                              GFP_NOFS);
1608
1609         BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1610
1611         spin_lock_init(&fs_info->block_group_cache_lock);
1612         fs_info->block_group_cache_tree.rb_node = NULL;
1613
1614         extent_io_tree_init(&fs_info->freed_extents[0],
1615                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1616         extent_io_tree_init(&fs_info->freed_extents[1],
1617                              fs_info->btree_inode->i_mapping, GFP_NOFS);
1618         fs_info->pinned_extents = &fs_info->freed_extents[0];
1619         fs_info->do_barriers = 1;
1620
1621         BTRFS_I(fs_info->btree_inode)->root = tree_root;
1622         memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1623                sizeof(struct btrfs_key));
1624         insert_inode_hash(fs_info->btree_inode);
1625
1626         mutex_init(&fs_info->trans_mutex);
1627         mutex_init(&fs_info->ordered_operations_mutex);
1628         mutex_init(&fs_info->tree_log_mutex);
1629         mutex_init(&fs_info->drop_mutex);
1630         mutex_init(&fs_info->chunk_mutex);
1631         mutex_init(&fs_info->transaction_kthread_mutex);
1632         mutex_init(&fs_info->cleaner_mutex);
1633         mutex_init(&fs_info->volume_mutex);
1634         mutex_init(&fs_info->tree_reloc_mutex);
1635         init_rwsem(&fs_info->extent_commit_sem);
1636
1637         btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1638         btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1639
1640         init_waitqueue_head(&fs_info->transaction_throttle);
1641         init_waitqueue_head(&fs_info->transaction_wait);
1642         init_waitqueue_head(&fs_info->async_submit_wait);
1643
1644         __setup_root(4096, 4096, 4096, 4096, tree_root,
1645                      fs_info, BTRFS_ROOT_TREE_OBJECTID);
1646
1647
1648         bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1649         if (!bh)
1650                 goto fail_iput;
1651
1652         memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1653         memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1654                sizeof(fs_info->super_for_commit));
1655         brelse(bh);
1656
1657         memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1658
1659         disk_super = &fs_info->super_copy;
1660         if (!btrfs_super_root(disk_super))
1661                 goto fail_iput;
1662
1663         ret = btrfs_parse_options(tree_root, options);
1664         if (ret) {
1665                 err = ret;
1666                 goto fail_iput;
1667         }
1668
1669         features = btrfs_super_incompat_flags(disk_super) &
1670                 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1671         if (features) {
1672                 printk(KERN_ERR "BTRFS: couldn't mount because of "
1673                        "unsupported optional features (%Lx).\n",
1674                        (unsigned long long)features);
1675                 err = -EINVAL;
1676                 goto fail_iput;
1677         }
1678
1679         features = btrfs_super_incompat_flags(disk_super);
1680         if (!(features & BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF)) {
1681                 features |= BTRFS_FEATURE_INCOMPAT_MIXED_BACKREF;
1682                 btrfs_set_super_incompat_flags(disk_super, features);
1683         }
1684
1685         features = btrfs_super_compat_ro_flags(disk_super) &
1686                 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1687         if (!(sb->s_flags & MS_RDONLY) && features) {
1688                 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1689                        "unsupported option features (%Lx).\n",
1690                        (unsigned long long)features);
1691                 err = -EINVAL;
1692                 goto fail_iput;
1693         }
1694 printk("thread pool is %d\n", fs_info->thread_pool_size);
1695         /*
1696          * we need to start all the end_io workers up front because the
1697          * queue work function gets called at interrupt time, and so it
1698          * cannot dynamically grow.
1699          */
1700         btrfs_init_workers(&fs_info->workers, "worker",
1701                            fs_info->thread_pool_size);
1702
1703         btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1704                            fs_info->thread_pool_size);
1705
1706         btrfs_init_workers(&fs_info->submit_workers, "submit",
1707                            min_t(u64, fs_devices->num_devices,
1708                            fs_info->thread_pool_size));
1709
1710         /* a higher idle thresh on the submit workers makes it much more
1711          * likely that bios will be send down in a sane order to the
1712          * devices
1713          */
1714         fs_info->submit_workers.idle_thresh = 64;
1715
1716         fs_info->workers.idle_thresh = 16;
1717         fs_info->workers.ordered = 1;
1718
1719         fs_info->delalloc_workers.idle_thresh = 2;
1720         fs_info->delalloc_workers.ordered = 1;
1721
1722         btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1723         btrfs_init_workers(&fs_info->endio_workers, "endio",
1724                            fs_info->thread_pool_size);
1725         btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1726                            fs_info->thread_pool_size);
1727         btrfs_init_workers(&fs_info->endio_meta_write_workers,
1728                            "endio-meta-write", fs_info->thread_pool_size);
1729         btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1730                            fs_info->thread_pool_size);
1731
1732         /*
1733          * endios are largely parallel and should have a very
1734          * low idle thresh
1735          */
1736         fs_info->endio_workers.idle_thresh = 4;
1737         fs_info->endio_meta_workers.idle_thresh = 4;
1738
1739         fs_info->endio_write_workers.idle_thresh = 2;
1740         fs_info->endio_meta_write_workers.idle_thresh = 2;
1741
1742         fs_info->endio_workers.atomic_worker_start = 1;
1743         fs_info->endio_meta_workers.atomic_worker_start = 1;
1744         fs_info->endio_write_workers.atomic_worker_start = 1;
1745         fs_info->endio_meta_write_workers.atomic_worker_start = 1;
1746
1747         btrfs_start_workers(&fs_info->workers, 1);
1748         btrfs_start_workers(&fs_info->submit_workers, 1);
1749         btrfs_start_workers(&fs_info->delalloc_workers, 1);
1750         btrfs_start_workers(&fs_info->fixup_workers, 1);
1751         btrfs_start_workers(&fs_info->endio_workers, 1);
1752         btrfs_start_workers(&fs_info->endio_meta_workers, 1);
1753         btrfs_start_workers(&fs_info->endio_meta_write_workers, 1);
1754         btrfs_start_workers(&fs_info->endio_write_workers, 1);
1755
1756         fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1757         fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1758                                     4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1759
1760         nodesize = btrfs_super_nodesize(disk_super);
1761         leafsize = btrfs_super_leafsize(disk_super);
1762         sectorsize = btrfs_super_sectorsize(disk_super);
1763         stripesize = btrfs_super_stripesize(disk_super);
1764         tree_root->nodesize = nodesize;
1765         tree_root->leafsize = leafsize;
1766         tree_root->sectorsize = sectorsize;
1767         tree_root->stripesize = stripesize;
1768
1769         sb->s_blocksize = sectorsize;
1770         sb->s_blocksize_bits = blksize_bits(sectorsize);
1771
1772         if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1773                     sizeof(disk_super->magic))) {
1774                 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1775                 goto fail_sb_buffer;
1776         }
1777
1778         mutex_lock(&fs_info->chunk_mutex);
1779         ret = btrfs_read_sys_array(tree_root);
1780         mutex_unlock(&fs_info->chunk_mutex);
1781         if (ret) {
1782                 printk(KERN_WARNING "btrfs: failed to read the system "
1783                        "array on %s\n", sb->s_id);
1784                 goto fail_sb_buffer;
1785         }
1786
1787         blocksize = btrfs_level_size(tree_root,
1788                                      btrfs_super_chunk_root_level(disk_super));
1789         generation = btrfs_super_chunk_root_generation(disk_super);
1790
1791         __setup_root(nodesize, leafsize, sectorsize, stripesize,
1792                      chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1793
1794         chunk_root->node = read_tree_block(chunk_root,
1795                                            btrfs_super_chunk_root(disk_super),
1796                                            blocksize, generation);
1797         BUG_ON(!chunk_root->node);
1798         if (!test_bit(EXTENT_BUFFER_UPTODATE, &chunk_root->node->bflags)) {
1799                 printk(KERN_WARNING "btrfs: failed to read chunk root on %s\n",
1800                        sb->s_id);
1801                 goto fail_chunk_root;
1802         }
1803         btrfs_set_root_node(&chunk_root->root_item, chunk_root->node);
1804         chunk_root->commit_root = btrfs_root_node(chunk_root);
1805
1806         read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1807            (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1808            BTRFS_UUID_SIZE);
1809
1810         mutex_lock(&fs_info->chunk_mutex);
1811         ret = btrfs_read_chunk_tree(chunk_root);
1812         mutex_unlock(&fs_info->chunk_mutex);
1813         if (ret) {
1814                 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1815                        sb->s_id);
1816                 goto fail_chunk_root;
1817         }
1818
1819         btrfs_close_extra_devices(fs_devices);
1820
1821         blocksize = btrfs_level_size(tree_root,
1822                                      btrfs_super_root_level(disk_super));
1823         generation = btrfs_super_generation(disk_super);
1824
1825         tree_root->node = read_tree_block(tree_root,
1826                                           btrfs_super_root(disk_super),
1827                                           blocksize, generation);
1828         if (!tree_root->node)
1829                 goto fail_chunk_root;
1830         if (!test_bit(EXTENT_BUFFER_UPTODATE, &tree_root->node->bflags)) {
1831                 printk(KERN_WARNING "btrfs: failed to read tree root on %s\n",
1832                        sb->s_id);
1833                 goto fail_tree_root;
1834         }
1835         btrfs_set_root_node(&tree_root->root_item, tree_root->node);
1836         tree_root->commit_root = btrfs_root_node(tree_root);
1837
1838         ret = find_and_setup_root(tree_root, fs_info,
1839                                   BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1840         if (ret)
1841                 goto fail_tree_root;
1842         extent_root->track_dirty = 1;
1843
1844         ret = find_and_setup_root(tree_root, fs_info,
1845                                   BTRFS_DEV_TREE_OBJECTID, dev_root);
1846         if (ret)
1847                 goto fail_extent_root;
1848         dev_root->track_dirty = 1;
1849
1850         ret = find_and_setup_root(tree_root, fs_info,
1851                                   BTRFS_CSUM_TREE_OBJECTID, csum_root);
1852         if (ret)
1853                 goto fail_dev_root;
1854
1855         csum_root->track_dirty = 1;
1856
1857         btrfs_read_block_groups(extent_root);
1858
1859         fs_info->generation = generation;
1860         fs_info->last_trans_committed = generation;
1861         fs_info->data_alloc_profile = (u64)-1;
1862         fs_info->metadata_alloc_profile = (u64)-1;
1863         fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1864         fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1865                                                "btrfs-cleaner");
1866         if (IS_ERR(fs_info->cleaner_kthread))
1867                 goto fail_csum_root;
1868
1869         fs_info->transaction_kthread = kthread_run(transaction_kthread,
1870                                                    tree_root,
1871                                                    "btrfs-transaction");
1872         if (IS_ERR(fs_info->transaction_kthread))
1873                 goto fail_cleaner;
1874
1875         if (!btrfs_test_opt(tree_root, SSD) &&
1876             !btrfs_test_opt(tree_root, NOSSD) &&
1877             !fs_info->fs_devices->rotating) {
1878                 printk(KERN_INFO "Btrfs detected SSD devices, enabling SSD "
1879                        "mode\n");
1880                 btrfs_set_opt(fs_info->mount_opt, SSD);
1881         }
1882
1883         if (btrfs_super_log_root(disk_super) != 0) {
1884                 u64 bytenr = btrfs_super_log_root(disk_super);
1885
1886                 if (fs_devices->rw_devices == 0) {
1887                         printk(KERN_WARNING "Btrfs log replay required "
1888                                "on RO media\n");
1889                         err = -EIO;
1890                         goto fail_trans_kthread;
1891                 }
1892                 blocksize =
1893                      btrfs_level_size(tree_root,
1894                                       btrfs_super_log_root_level(disk_super));
1895
1896                 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1897                                                       GFP_NOFS);
1898
1899                 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1900                              log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1901
1902                 log_tree_root->node = read_tree_block(tree_root, bytenr,
1903                                                       blocksize,
1904                                                       generation + 1);
1905                 ret = btrfs_recover_log_trees(log_tree_root);
1906                 BUG_ON(ret);
1907
1908                 if (sb->s_flags & MS_RDONLY) {
1909                         ret =  btrfs_commit_super(tree_root);
1910                         BUG_ON(ret);
1911                 }
1912         }
1913
1914         if (!(sb->s_flags & MS_RDONLY)) {
1915                 ret = btrfs_recover_relocation(tree_root);
1916                 BUG_ON(ret);
1917         }
1918
1919         location.objectid = BTRFS_FS_TREE_OBJECTID;
1920         location.type = BTRFS_ROOT_ITEM_KEY;
1921         location.offset = (u64)-1;
1922
1923         fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1924         if (!fs_info->fs_root)
1925                 goto fail_trans_kthread;
1926
1927         return tree_root;
1928
1929 fail_trans_kthread:
1930         kthread_stop(fs_info->transaction_kthread);
1931 fail_cleaner:
1932         kthread_stop(fs_info->cleaner_kthread);
1933
1934         /*
1935          * make sure we're done with the btree inode before we stop our
1936          * kthreads
1937          */
1938         filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1939         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1940
1941 fail_csum_root:
1942         free_extent_buffer(csum_root->node);
1943         free_extent_buffer(csum_root->commit_root);
1944 fail_dev_root:
1945         free_extent_buffer(dev_root->node);
1946         free_extent_buffer(dev_root->commit_root);
1947 fail_extent_root:
1948         free_extent_buffer(extent_root->node);
1949         free_extent_buffer(extent_root->commit_root);
1950 fail_tree_root:
1951         free_extent_buffer(tree_root->node);
1952         free_extent_buffer(tree_root->commit_root);
1953 fail_chunk_root:
1954         free_extent_buffer(chunk_root->node);
1955         free_extent_buffer(chunk_root->commit_root);
1956 fail_sb_buffer:
1957         btrfs_stop_workers(&fs_info->fixup_workers);
1958         btrfs_stop_workers(&fs_info->delalloc_workers);
1959         btrfs_stop_workers(&fs_info->workers);
1960         btrfs_stop_workers(&fs_info->endio_workers);
1961         btrfs_stop_workers(&fs_info->endio_meta_workers);
1962         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1963         btrfs_stop_workers(&fs_info->endio_write_workers);
1964         btrfs_stop_workers(&fs_info->submit_workers);
1965 fail_iput:
1966         invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1967         iput(fs_info->btree_inode);
1968
1969         btrfs_close_devices(fs_info->fs_devices);
1970         btrfs_mapping_tree_free(&fs_info->mapping_tree);
1971 fail_bdi:
1972         bdi_destroy(&fs_info->bdi);
1973 fail:
1974         kfree(extent_root);
1975         kfree(tree_root);
1976         kfree(fs_info);
1977         kfree(chunk_root);
1978         kfree(dev_root);
1979         kfree(csum_root);
1980         return ERR_PTR(err);
1981 }
1982
1983 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1984 {
1985         char b[BDEVNAME_SIZE];
1986
1987         if (uptodate) {
1988                 set_buffer_uptodate(bh);
1989         } else {
1990                 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1991                         printk(KERN_WARNING "lost page write due to "
1992                                         "I/O error on %s\n",
1993                                        bdevname(bh->b_bdev, b));
1994                 }
1995                 /* note, we dont' set_buffer_write_io_error because we have
1996                  * our own ways of dealing with the IO errors
1997                  */
1998                 clear_buffer_uptodate(bh);
1999         }
2000         unlock_buffer(bh);
2001         put_bh(bh);
2002 }
2003
2004 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
2005 {
2006         struct buffer_head *bh;
2007         struct buffer_head *latest = NULL;
2008         struct btrfs_super_block *super;
2009         int i;
2010         u64 transid = 0;
2011         u64 bytenr;
2012
2013         /* we would like to check all the supers, but that would make
2014          * a btrfs mount succeed after a mkfs from a different FS.
2015          * So, we need to add a special mount option to scan for
2016          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2017          */
2018         for (i = 0; i < 1; i++) {
2019                 bytenr = btrfs_sb_offset(i);
2020                 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2021                         break;
2022                 bh = __bread(bdev, bytenr / 4096, 4096);
2023                 if (!bh)
2024                         continue;
2025
2026                 super = (struct btrfs_super_block *)bh->b_data;
2027                 if (btrfs_super_bytenr(super) != bytenr ||
2028                     strncmp((char *)(&super->magic), BTRFS_MAGIC,
2029                             sizeof(super->magic))) {
2030                         brelse(bh);
2031                         continue;
2032                 }
2033
2034                 if (!latest || btrfs_super_generation(super) > transid) {
2035                         brelse(latest);
2036                         latest = bh;
2037                         transid = btrfs_super_generation(super);
2038                 } else {
2039                         brelse(bh);
2040                 }
2041         }
2042         return latest;
2043 }
2044
2045 /*
2046  * this should be called twice, once with wait == 0 and
2047  * once with wait == 1.  When wait == 0 is done, all the buffer heads
2048  * we write are pinned.
2049  *
2050  * They are released when wait == 1 is done.
2051  * max_mirrors must be the same for both runs, and it indicates how
2052  * many supers on this one device should be written.
2053  *
2054  * max_mirrors == 0 means to write them all.
2055  */
2056 static int write_dev_supers(struct btrfs_device *device,
2057                             struct btrfs_super_block *sb,
2058                             int do_barriers, int wait, int max_mirrors)
2059 {
2060         struct buffer_head *bh;
2061         int i;
2062         int ret;
2063         int errors = 0;
2064         u32 crc;
2065         u64 bytenr;
2066         int last_barrier = 0;
2067
2068         if (max_mirrors == 0)
2069                 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2070
2071         /* make sure only the last submit_bh does a barrier */
2072         if (do_barriers) {
2073                 for (i = 0; i < max_mirrors; i++) {
2074                         bytenr = btrfs_sb_offset(i);
2075                         if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2076                             device->total_bytes)
2077                                 break;
2078                         last_barrier = i;
2079                 }
2080         }
2081
2082         for (i = 0; i < max_mirrors; i++) {
2083                 bytenr = btrfs_sb_offset(i);
2084                 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2085                         break;
2086
2087                 if (wait) {
2088                         bh = __find_get_block(device->bdev, bytenr / 4096,
2089                                               BTRFS_SUPER_INFO_SIZE);
2090                         BUG_ON(!bh);
2091                         wait_on_buffer(bh);
2092                         if (!buffer_uptodate(bh))
2093                                 errors++;
2094
2095                         /* drop our reference */
2096                         brelse(bh);
2097
2098                         /* drop the reference from the wait == 0 run */
2099                         brelse(bh);
2100                         continue;
2101                 } else {
2102                         btrfs_set_super_bytenr(sb, bytenr);
2103
2104                         crc = ~(u32)0;
2105                         crc = btrfs_csum_data(NULL, (char *)sb +
2106                                               BTRFS_CSUM_SIZE, crc,
2107                                               BTRFS_SUPER_INFO_SIZE -
2108                                               BTRFS_CSUM_SIZE);
2109                         btrfs_csum_final(crc, sb->csum);
2110
2111                         /*
2112                          * one reference for us, and we leave it for the
2113                          * caller
2114                          */
2115                         bh = __getblk(device->bdev, bytenr / 4096,
2116                                       BTRFS_SUPER_INFO_SIZE);
2117                         memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2118
2119                         /* one reference for submit_bh */
2120                         get_bh(bh);
2121
2122                         set_buffer_uptodate(bh);
2123                         lock_buffer(bh);
2124                         bh->b_end_io = btrfs_end_buffer_write_sync;
2125                 }
2126
2127                 if (i == last_barrier && do_barriers && device->barriers) {
2128                         ret = submit_bh(WRITE_BARRIER, bh);
2129                         if (ret == -EOPNOTSUPP) {
2130                                 printk("btrfs: disabling barriers on dev %s\n",
2131                                        device->name);
2132                                 set_buffer_uptodate(bh);
2133                                 device->barriers = 0;
2134                                 /* one reference for submit_bh */
2135                                 get_bh(bh);
2136                                 lock_buffer(bh);
2137                                 ret = submit_bh(WRITE_SYNC, bh);
2138                         }
2139                 } else {
2140                         ret = submit_bh(WRITE_SYNC, bh);
2141                 }
2142
2143                 if (ret)
2144                         errors++;
2145         }
2146         return errors < i ? 0 : -1;
2147 }
2148
2149 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2150 {
2151         struct list_head *head;
2152         struct btrfs_device *dev;
2153         struct btrfs_super_block *sb;
2154         struct btrfs_dev_item *dev_item;
2155         int ret;
2156         int do_barriers;
2157         int max_errors;
2158         int total_errors = 0;
2159         u64 flags;
2160
2161         max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2162         do_barriers = !btrfs_test_opt(root, NOBARRIER);
2163
2164         sb = &root->fs_info->super_for_commit;
2165         dev_item = &sb->dev_item;
2166
2167         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2168         head = &root->fs_info->fs_devices->devices;
2169         list_for_each_entry(dev, head, dev_list) {
2170                 if (!dev->bdev) {
2171                         total_errors++;
2172                         continue;
2173                 }
2174                 if (!dev->in_fs_metadata || !dev->writeable)
2175                         continue;
2176
2177                 btrfs_set_stack_device_generation(dev_item, 0);
2178                 btrfs_set_stack_device_type(dev_item, dev->type);
2179                 btrfs_set_stack_device_id(dev_item, dev->devid);
2180                 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2181                 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2182                 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2183                 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2184                 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2185                 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2186                 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2187
2188                 flags = btrfs_super_flags(sb);
2189                 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2190
2191                 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2192                 if (ret)
2193                         total_errors++;
2194         }
2195         if (total_errors > max_errors) {
2196                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2197                        total_errors);
2198                 BUG();
2199         }
2200
2201         total_errors = 0;
2202         list_for_each_entry(dev, head, dev_list) {
2203                 if (!dev->bdev)
2204                         continue;
2205                 if (!dev->in_fs_metadata || !dev->writeable)
2206                         continue;
2207
2208                 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2209                 if (ret)
2210                         total_errors++;
2211         }
2212         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2213         if (total_errors > max_errors) {
2214                 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2215                        total_errors);
2216                 BUG();
2217         }
2218         return 0;
2219 }
2220
2221 int write_ctree_super(struct btrfs_trans_handle *trans,
2222                       struct btrfs_root *root, int max_mirrors)
2223 {
2224         int ret;
2225
2226         ret = write_all_supers(root, max_mirrors);
2227         return ret;
2228 }
2229
2230 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2231 {
2232         WARN_ON(!RB_EMPTY_ROOT(&root->inode_tree));
2233         radix_tree_delete(&fs_info->fs_roots_radix,
2234                           (unsigned long)root->root_key.objectid);
2235         if (root->anon_super.s_dev) {
2236                 down_write(&root->anon_super.s_umount);
2237                 kill_anon_super(&root->anon_super);
2238         }
2239         if (root->node)
2240                 free_extent_buffer(root->node);
2241         if (root->commit_root)
2242                 free_extent_buffer(root->commit_root);
2243         kfree(root->name);
2244         kfree(root);
2245         return 0;
2246 }
2247
2248 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2249 {
2250         int ret;
2251         struct btrfs_root *gang[8];
2252         int i;
2253
2254         while (1) {
2255                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2256                                              (void **)gang, 0,
2257                                              ARRAY_SIZE(gang));
2258                 if (!ret)
2259                         break;
2260                 for (i = 0; i < ret; i++)
2261                         btrfs_free_fs_root(fs_info, gang[i]);
2262         }
2263         return 0;
2264 }
2265
2266 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2267 {
2268         u64 root_objectid = 0;
2269         struct btrfs_root *gang[8];
2270         int i;
2271         int ret;
2272
2273         while (1) {
2274                 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2275                                              (void **)gang, root_objectid,
2276                                              ARRAY_SIZE(gang));
2277                 if (!ret)
2278                         break;
2279
2280                 root_objectid = gang[ret - 1]->root_key.objectid + 1;
2281                 for (i = 0; i < ret; i++) {
2282                         root_objectid = gang[i]->root_key.objectid;
2283                         ret = btrfs_find_dead_roots(fs_info->tree_root,
2284                                                     root_objectid);
2285                         BUG_ON(ret);
2286                         btrfs_orphan_cleanup(gang[i]);
2287                 }
2288                 root_objectid++;
2289         }
2290         return 0;
2291 }
2292
2293 int btrfs_commit_super(struct btrfs_root *root)
2294 {
2295         struct btrfs_trans_handle *trans;
2296         int ret;
2297
2298         mutex_lock(&root->fs_info->cleaner_mutex);
2299         btrfs_clean_old_snapshots(root);
2300         mutex_unlock(&root->fs_info->cleaner_mutex);
2301         trans = btrfs_start_transaction(root, 1);
2302         ret = btrfs_commit_transaction(trans, root);
2303         BUG_ON(ret);
2304         /* run commit again to drop the original snapshot */
2305         trans = btrfs_start_transaction(root, 1);
2306         btrfs_commit_transaction(trans, root);
2307         ret = btrfs_write_and_wait_transaction(NULL, root);
2308         BUG_ON(ret);
2309
2310         ret = write_ctree_super(NULL, root, 0);
2311         return ret;
2312 }
2313
2314 int close_ctree(struct btrfs_root *root)
2315 {
2316         struct btrfs_fs_info *fs_info = root->fs_info;
2317         int ret;
2318
2319         fs_info->closing = 1;
2320         smp_mb();
2321
2322         kthread_stop(root->fs_info->transaction_kthread);
2323         kthread_stop(root->fs_info->cleaner_kthread);
2324
2325         if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2326                 ret =  btrfs_commit_super(root);
2327                 if (ret)
2328                         printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2329         }
2330
2331         fs_info->closing = 2;
2332         smp_mb();
2333
2334         if (fs_info->delalloc_bytes) {
2335                 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2336                        (unsigned long long)fs_info->delalloc_bytes);
2337         }
2338         if (fs_info->total_ref_cache_size) {
2339                 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2340                        (unsigned long long)fs_info->total_ref_cache_size);
2341         }
2342
2343         free_extent_buffer(fs_info->extent_root->node);
2344         free_extent_buffer(fs_info->extent_root->commit_root);
2345         free_extent_buffer(fs_info->tree_root->node);
2346         free_extent_buffer(fs_info->tree_root->commit_root);
2347         free_extent_buffer(root->fs_info->chunk_root->node);
2348         free_extent_buffer(root->fs_info->chunk_root->commit_root);
2349         free_extent_buffer(root->fs_info->dev_root->node);
2350         free_extent_buffer(root->fs_info->dev_root->commit_root);
2351         free_extent_buffer(root->fs_info->csum_root->node);
2352         free_extent_buffer(root->fs_info->csum_root->commit_root);
2353
2354         btrfs_free_block_groups(root->fs_info);
2355
2356         del_fs_roots(fs_info);
2357
2358         iput(fs_info->btree_inode);
2359
2360         btrfs_stop_workers(&fs_info->fixup_workers);
2361         btrfs_stop_workers(&fs_info->delalloc_workers);
2362         btrfs_stop_workers(&fs_info->workers);
2363         btrfs_stop_workers(&fs_info->endio_workers);
2364         btrfs_stop_workers(&fs_info->endio_meta_workers);
2365         btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2366         btrfs_stop_workers(&fs_info->endio_write_workers);
2367         btrfs_stop_workers(&fs_info->submit_workers);
2368
2369         btrfs_close_devices(fs_info->fs_devices);
2370         btrfs_mapping_tree_free(&fs_info->mapping_tree);
2371
2372         bdi_destroy(&fs_info->bdi);
2373
2374         kfree(fs_info->extent_root);
2375         kfree(fs_info->tree_root);
2376         kfree(fs_info->chunk_root);
2377         kfree(fs_info->dev_root);
2378         kfree(fs_info->csum_root);
2379         return 0;
2380 }
2381
2382 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2383 {
2384         int ret;
2385         struct inode *btree_inode = buf->first_page->mapping->host;
2386
2387         ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2388         if (!ret)
2389                 return ret;
2390
2391         ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2392                                     parent_transid);
2393         return !ret;
2394 }
2395
2396 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2397 {
2398         struct inode *btree_inode = buf->first_page->mapping->host;
2399         return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2400                                           buf);
2401 }
2402
2403 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2404 {
2405         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2406         u64 transid = btrfs_header_generation(buf);
2407         struct inode *btree_inode = root->fs_info->btree_inode;
2408         int was_dirty;
2409
2410         btrfs_assert_tree_locked(buf);
2411         if (transid != root->fs_info->generation) {
2412                 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2413                        "found %llu running %llu\n",
2414                         (unsigned long long)buf->start,
2415                         (unsigned long long)transid,
2416                         (unsigned long long)root->fs_info->generation);
2417                 WARN_ON(1);
2418         }
2419         was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2420                                             buf);
2421         if (!was_dirty) {
2422                 spin_lock(&root->fs_info->delalloc_lock);
2423                 root->fs_info->dirty_metadata_bytes += buf->len;
2424                 spin_unlock(&root->fs_info->delalloc_lock);
2425         }
2426 }
2427
2428 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2429 {
2430         /*
2431          * looks as though older kernels can get into trouble with
2432          * this code, they end up stuck in balance_dirty_pages forever
2433          */
2434         u64 num_dirty;
2435         unsigned long thresh = 32 * 1024 * 1024;
2436
2437         if (current->flags & PF_MEMALLOC)
2438                 return;
2439
2440         num_dirty = root->fs_info->dirty_metadata_bytes;
2441
2442         if (num_dirty > thresh) {
2443                 balance_dirty_pages_ratelimited_nr(
2444                                    root->fs_info->btree_inode->i_mapping, 1);
2445         }
2446         return;
2447 }
2448
2449 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2450 {
2451         struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2452         int ret;
2453         ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2454         if (ret == 0)
2455                 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2456         return ret;
2457 }
2458
2459 int btree_lock_page_hook(struct page *page)
2460 {
2461         struct inode *inode = page->mapping->host;
2462         struct btrfs_root *root = BTRFS_I(inode)->root;
2463         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2464         struct extent_buffer *eb;
2465         unsigned long len;
2466         u64 bytenr = page_offset(page);
2467
2468         if (page->private == EXTENT_PAGE_PRIVATE)
2469                 goto out;
2470
2471         len = page->private >> 2;
2472         eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2473         if (!eb)
2474                 goto out;
2475
2476         btrfs_tree_lock(eb);
2477         btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2478
2479         if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2480                 spin_lock(&root->fs_info->delalloc_lock);
2481                 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2482                         root->fs_info->dirty_metadata_bytes -= eb->len;
2483                 else
2484                         WARN_ON(1);
2485                 spin_unlock(&root->fs_info->delalloc_lock);
2486         }
2487
2488         btrfs_tree_unlock(eb);
2489         free_extent_buffer(eb);
2490 out:
2491         lock_page(page);
2492         return 0;
2493 }
2494
2495 static struct extent_io_ops btree_extent_io_ops = {
2496         .write_cache_pages_lock_hook = btree_lock_page_hook,
2497         .readpage_end_io_hook = btree_readpage_end_io_hook,
2498         .submit_bio_hook = btree_submit_bio_hook,
2499         /* note we're sharing with inode.c for the merge bio hook */
2500         .merge_bio_hook = btrfs_merge_bio_hook,
2501 };